1 /* 2 * Copyright 2011-2015 Samy Al Bahra. 3 * Copyright 2011 David Joseph. 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <ck_barrier.h> 29 #include <ck_cc.h> 30 #include <ck_pr.h> 31 #include <ck_spinlock.h> 32 33 #include "ck_internal.h" 34 35 void 36 ck_barrier_dissemination_init(struct ck_barrier_dissemination *barrier, 37 struct ck_barrier_dissemination_flag **barrier_internal, 38 unsigned int nthr) 39 { 40 unsigned int i, j, k, size, offset; 41 bool p = nthr & (nthr - 1); 42 43 barrier->nthr = nthr; 44 barrier->size = size = ck_internal_log(ck_internal_power_2(nthr)); 45 ck_pr_store_uint(&barrier->tid, 0); 46 47 for (i = 0; i < nthr; ++i) { 48 barrier[i].flags[0] = barrier_internal[i]; 49 barrier[i].flags[1] = barrier_internal[i] + size; 50 } 51 52 for (i = 0; i < nthr; ++i) { 53 for (k = 0, offset = 1; k < size; ++k, offset <<= 1) { 54 /* 55 * Determine the thread's partner, j, for the current round, k. 56 * Partners are chosen such that by the completion of the barrier, 57 * every thread has been directly (having one of its flag set) or 58 * indirectly (having one of its partners's flags set) signaled 59 * by every other thread in the barrier. 60 */ 61 if (p == false) 62 j = (i + offset) & (nthr - 1); 63 else 64 j = (i + offset) % nthr; 65 66 /* Set the thread's partner for round k. */ 67 barrier[i].flags[0][k].pflag = &barrier[j].flags[0][k].tflag; 68 barrier[i].flags[1][k].pflag = &barrier[j].flags[1][k].tflag; 69 70 /* Set the thread's flags to false. */ 71 barrier[i].flags[0][k].tflag = barrier[i].flags[1][k].tflag = 0; 72 } 73 } 74 75 return; 76 } 77 78 void 79 ck_barrier_dissemination_subscribe(struct ck_barrier_dissemination *barrier, 80 struct ck_barrier_dissemination_state *state) 81 { 82 83 state->parity = 0; 84 state->sense = ~0; 85 state->tid = ck_pr_faa_uint(&barrier->tid, 1); 86 return; 87 } 88 89 unsigned int 90 ck_barrier_dissemination_size(unsigned int nthr) 91 { 92 93 return (ck_internal_log(ck_internal_power_2(nthr)) << 1); 94 } 95 96 void 97 ck_barrier_dissemination(struct ck_barrier_dissemination *barrier, 98 struct ck_barrier_dissemination_state *state) 99 { 100 unsigned int i; 101 unsigned int size = barrier->size; 102 103 for (i = 0; i < size; ++i) { 104 unsigned int *pflag, *tflag; 105 106 pflag = barrier[state->tid].flags[state->parity][i].pflag; 107 tflag = &barrier[state->tid].flags[state->parity][i].tflag; 108 109 /* Unblock current partner. */ 110 ck_pr_store_uint(pflag, state->sense); 111 112 /* Wait until some other thread unblocks this one. */ 113 while (ck_pr_load_uint(tflag) != state->sense) 114 ck_pr_stall(); 115 } 116 117 /* 118 * Dissemination barriers use two sets of flags to prevent race conditions 119 * between successive calls to the barrier. Parity indicates which set will 120 * be used for the next barrier. They also use a sense reversal technique 121 * to avoid re-initialization of the flags for every two calls to the barrier. 122 */ 123 if (state->parity == 1) 124 state->sense = ~state->sense; 125 126 state->parity = 1 - state->parity; 127 128 ck_pr_fence_acquire(); 129 return; 130 } 131